Type Ia supernovae are generally thought to be due to the thermonuclear explosions of carbon-oxygen white dwarfs with masses near the Chandrasekhar mass. This scenario, however, has two long-standing problems. First, the explosions do not naturally produce the correct mix of elements, but have to be finely tuned to proceed from sub-sonic deflagration to super-sonic detonation. Second, population models and observations give formation rates of near-Chandrasekhar white dwarfs that are far too small. Here, we suggest that type Ia supernovae instead result from mergers of roughly equal-mass carbon-oxygen white dwarfs, including those that produce sub-Chandrasekhar mass remnants. Numerical studies of such mergers have shown that the remnants consist of rapidly rotating cores that contain most of the mass and are hottest in the center, surrounded by dense, small disks. We argue that the disks accrete quickly, and that the resulting compressional heating likely leads to central carbon ignition. This ignition occurs at densities for which pure detonations lead to events similar to type Ia supernovae. With this merger scenario, we can understand the type Ia rates, and have plausible reasons for the observed range in luminosity and for the bias of more luminous supernovae towards younger populations. We speculate that explosions of white dwarfs slowly brought to the Chandrasekhar limit---which should also occur---are responsible for some of the "atypical" type Ia supernovae.